An AI-first notebook whose whole pitch is removing the filing decision: "don't decide where it goes, throw it in, ask for it later." A Chrome extension auto-summarizes saved content, an AI layer auto-tags and auto-links what comes in, and "Mem Chat" queries across your accumulated history.
Minimizing capture friction to nearly zero by fully automating organization — the same instinct behind Mimir's own "drop it in, sort it never" fast-lane capture, running for years as a shipped consumer product rather than a design document.
Both explicitly avoid manual filing at capture time and both apply automated tagging on top of raw capture — Mem's AI auto-tagging versus Mimir's deterministic-first facet inheritance, which is a genuinely different mechanism aimed at the same problem.
Mimir's tagging is deterministic-first by design specifically to avoid the taxonomy drift an LLM-per-atom approach invites at real scale — one ingestion run tagging client/SteadyStars and another tagging client/ss silently becomes two facets instead of one. Mimir also adds a compulsory governance layer, a formal chunk-to-source provenance chain, and row-level classification tiers — none of which the zero-structure/RAG-first school Mem belongs to has an equivalent of.
Mem is a shipped, polished consumer product working for real users today. Its exact retrieval architecture isn't independently documented in the research this site draws on — worth stating honestly rather than guessing — but "opaque and working" is still ahead of a from-scratch system that isn't fully running yet. Mem also needs zero setup, versus Mimir's Postgres-plus-serving-daemon requirement.
research/07-zero-structure-rag.md · research/_tools-and-stacks.md